# CoreWeave Docs: Japanese インフラストラクチャー ワークロード > Learn how to deploy, manage, and observe your AI training and inference workloads on CoreWeave. ## ワークロード - [Workloads](https://docs.coreweave.com/products/workloads-home.md): Overview of SUNK, CoreWeave Inference, and CoreWeave Sandbox for running AI workloads on CoreWeave ### SUNK - [About SUNK](https://docs.coreweave.com/products/sunk.md): Run Slurm on Kubernetes with SUNK, in the SUNK Standard or SUNK Self-Service operation mode - [Choose a SUNK operation mode](https://docs.coreweave.com/products/sunk/choose-operation-mode.md): Pick SUNK Standard or SUNK Self-Service based on who operates the control plane and what you need to configure - [About SUNK Self-Service](https://docs.coreweave.com/products/sunk/about-self-service-sunk.md): The operation mode in which CoreWeave runs SUNK for you on CKS. Declare a Slurm cluster as one SunkCluster resource. #### 概念 - [Compute and login nodes](https://docs.coreweave.com/products/sunk/discover_sunk/compute_and_login_nodes.md): Understand the roles of SUNK compute nodes and login nodes as Kubernetes Pods running Slurm - [Three-layer orchestration model](https://docs.coreweave.com/products/sunk/discover_sunk/three-layers-of-orchestration.md): Understand CoreWeave's three-layer orchestration model - [SUNK Self-Service cluster lifecycle and reconciliation](https://docs.coreweave.com/products/sunk/discover_sunk/cluster-lifecycle-and-reconciliation.md): How a SUNK Self-Service cluster is created and reconciled, and how to read its status conditions to track progress - [SUNK Self-Service architecture](https://docs.coreweave.com/products/sunk/discover_sunk/self-service-architecture.md): The two operators behind SUNK Self-Service: where each runs, what it watches, and what it creates - [NodeSet](https://docs.coreweave.com/products/sunk/discover_sunk/nodeset.md): The NodeSet CRD that represents a set of Slurm nodes as Kubernetes Pods with 1:1 Node mapping - [Syncer](https://docs.coreweave.com/products/sunk/discover_sunk/syncer.md): How the SUNK Syncer bidirectionally synchronizes state between Slurm nodes and Kubernetes Pods - [SUNK Pod Scheduler](https://docs.coreweave.com/products/sunk/discover_sunk/scheduler.md): How the SUNK Pod Scheduler uses Slurm's scheduling mechanism to schedule Kubernetes Pods - [Node Controller](https://docs.coreweave.com/products/sunk/discover_sunk/node-controller.md): How the SUNK Node Controller synchronizes data between Kubernetes Nodes and slurmd Pods - [Pod Controller](https://docs.coreweave.com/products/sunk/discover_sunk/pod-controller.md): How the SUNK Pod Controller synchronizes data from Kubernetes Nodes to Pods for cordon and lock state - [Node locking](https://docs.coreweave.com/products/sunk/discover_sunk/node-locking.md): How SUNK locks Kubernetes Nodes to prevent other workloads during active Slurm job execution - [SlurmCluster](https://docs.coreweave.com/products/sunk/discover_sunk/slurmcluster.md): The SlurmCluster CRD that configures cluster-wide Slurm information and generates topology files - [Introduction to GPU straggler detection](https://docs.coreweave.com/products/sunk/discover_sunk/straggler-detection.md): Learn about CoreWeave Mission Control™ GPU Straggler Detection - [Fault tolerance in SUNK](https://docs.coreweave.com/products/sunk/discover_sunk/fault-tolerance.md): How CoreWeave removes failing Nodes and how Slurm requeue restarts your SUNK jobs after a failure - [ML container images](https://docs.coreweave.com/products/sunk/discover_sunk/ml-containers.md): Overview of CoreWeave's optimized PyTorch container images for distributed training on CKS and SUNK - [Choose and troubleshoot a training container image](https://docs.coreweave.com/products/sunk/discover_sunk/choosing-training-images.md): Pick a production-safe ML container tag, match CUDA to your GPU driver, and diagnose image-related failures #### チュートリアル - [Get started with SUNK Self-Service using the SunkCluster CR](https://docs.coreweave.com/products/sunk/tutorials/get-started-with-sunkcluster-cr.md): Stand up a SUNK Self-Service cluster end to end by applying a SunkCluster custom resource to an existing CKS cluster - [Run GitHub Actions runners on SUNK](https://docs.coreweave.com/products/sunk/tutorials/github-actions-runners.md): Deploy self-hosted GitHub Actions runners using Actions Runner Controller (ARC) - [Run notebooks on SUNK](https://docs.coreweave.com/products/sunk/tutorials/notebooks-on-sunk.md): Run interactive marimo and Jupyter notebooks on SUNK clusters with GPU resources - [Profile Python applications on SUNK](https://docs.coreweave.com/products/sunk/tutorials/profiling-python-apps-in-kubernetes.md): Profile Python applications running in Slurm jobs using py-spy and Linux perf tools - [Run Ray on SUNK](https://docs.coreweave.com/products/sunk/tutorials/ray-on-sunk.md): Integrate Ray's distributed computing framework with SUNK's Slurm-based scheduler for parallel workloads - [Run SkyPilot on SUNK](https://docs.coreweave.com/products/sunk/tutorials/skypilot-on-sunk.md): Use SkyPilot to run containerized GPU workloads and interactive development sessions on SUNK - [Run SWE-bench in SUNK with Docker](https://docs.coreweave.com/products/sunk/tutorials/swe-bench-in-sunk.md): Run the SWE-bench LLM benchmark on SUNK using Docker-in-Docker with GPU resources - [Run torchforge on SUNK](https://docs.coreweave.com/products/sunk/tutorials/torchforge-on-sunk.md): Run Grouped Relative Policy Optimization (GRPO) training with torchforge on SUNK - [Run veRL on SUNK](https://docs.coreweave.com/products/sunk/tutorials/verl-on-sunk.md): Run Group Relative Policy Optimization (GRPO) training with veRL on SUNK ##### SUNK でトレーニングする - [Train on SUNK](https://docs.coreweave.com/products/sunk/tutorials/train-on-sunk.md): Multi-part tutorial for training a model on SUNK using Slurm, from cluster setup to job monitoring - [1. Set up a Slurm cluster](https://docs.coreweave.com/products/sunk/tutorials/train-on-sunk/1-set-up-slurm-cluster.md): Connect to a Slurm login node through SSH and verify cluster access to begin training on SUNK - [2. Submit a simple job to the cluster](https://docs.coreweave.com/products/sunk/tutorials/train-on-sunk/2-submit-simple-job.md): Prepare and submit a basic Slurm batch job script with SBATCH directives and environment variables - [3. Submit a training job with PyTorch or TensorFlow](https://docs.coreweave.com/products/sunk/tutorials/train-on-sunk/3-submit-a-training-job.md): Submit multi-node, multi-GPU distributed training jobs using PyTorch torchrun on Slurm - [4. Monitor jobs with Grafana](https://docs.coreweave.com/products/sunk/tutorials/train-on-sunk/4-monitor-jobs.md): Monitor Slurm training jobs using CoreWeave's Grafana dashboards for GPU utilization and performance metrics #### SUNK をデプロイする - [Create a SUNK Self-Service cluster](https://docs.coreweave.com/products/sunk/deploy_sunk/create-sunk-cluster.md): Create a SUNK Self-Service cluster on CKS using the Cloud Console or a SunkCluster custom resource - [Upgrade a SUNK Self-Service cluster](https://docs.coreweave.com/products/sunk/deploy_sunk/upgrade-sunk-cluster.md): The version fields that control a SUNK Self-Service cluster and what to expect when you upgrade the cluster - [Configure compute nodes for SUNK Standard](https://docs.coreweave.com/products/sunk/deploy_sunk/configure-compute-nodes.md): Define Slurm compute node resources in the Slurm Helm chart values for SUNK Standard - [Bind NodeSets to Node Pools for SUNK Standard](https://docs.coreweave.com/products/sunk/deploy_sunk/bind-nodesets-to-node-pools.md): Match Node Pool labels and taints to NodeSet affinity and tolerations so slurmd Pods land on the intended Nodes - [Configure individual login Pods](https://docs.coreweave.com/products/sunk/deploy_sunk/configure-slurm-individual-login.md): Configure per-user Slurm login Pods on SUNK 7.x and earlier through the slurm-login subchart - [Manage a SUNK Standard deployment with CI and GitOps](https://docs.coreweave.com/products/sunk/deploy_sunk/manage-deployment-with-ci.md): Manage a SUNK Standard deployment with ArgoCD and GitOps-based CI workflows #### SUNK Anywhere - [About SUNK Anywhere](https://docs.coreweave.com/products/sunk/sunk_anywhere/about.md): Run SUNK Standard under license on a Kubernetes cluster that CoreWeave doesn't operate, such as GKE or EKS - [Get started with SUNK Anywhere](https://docs.coreweave.com/products/sunk/sunk_anywhere/get-started.md): Pull the SUNK operator image and Helm charts with a CoreWeave API access token, then deploy SUNK on your own cluster #### SUNK にアクセスする - [Connect to the Slurm login node](https://docs.coreweave.com/products/sunk/access_sunk/connect-to-slurm-login-node.md): Connect to an individual Slurm login node using SSH - [Access Slurm compute nodes](https://docs.coreweave.com/products/sunk/access_sunk/access-slurm-compute-nodes.md): Open a shell on a compute node, attach to a job's running container, or connect over SSH through the login node - [Tunnel VS Code for development on Slurm](https://docs.coreweave.com/products/sunk/access_sunk/vs-code-with-slurm.md): Use VS Code tunnels with SUNK to develop on compute nodes #### SUNK を管理する - [Interact with SUNK using kubectl](https://docs.coreweave.com/products/sunk/manage_sunk/interact-with-kubernetes.md): Use kubectl to interact with SUNK Kubernetes resources, access logs, and restart the Slurm Controller ##### ユーザーとアクセス - [Provision users in SUNK](https://docs.coreweave.com/products/sunk/manage_sunk/manage_cluster_access/sunk_user_provisioning.md): Set up SUNK User Provisioning (SUP) with CoreWeave IAM - [Manage users with nsscache on SUNK Standard](https://docs.coreweave.com/products/sunk/manage_sunk/manage_cluster_access/nsscache.md): Enable automatic user provisioning with nsscache on a SUNK Standard cluster - [Manage users with a directory service on SUNK Standard](https://docs.coreweave.com/products/sunk/manage_sunk/manage_cluster_access/directory-services.md): Manage user accounts on a SUNK Standard cluster with an external directory service - [Restrict compute node access with pam_slurm_adopt](https://docs.coreweave.com/products/sunk/manage_sunk/manage_cluster_access/pam-slurm-adopt.md): Configure the pam_slurm_adopt PAM module to restrict SSH access to Slurm compute nodes to users with active job allocations ##### ストレージ - [Configure storage mounts on SUNK Self-Service](https://docs.coreweave.com/products/sunk/manage_sunk/configure-storage-mounts.md): Size the home directory and add shared storage mounts on a SUNK Self-Service cluster through spec.storage - [Share storage across Slurm nodes on SUNK Standard](https://docs.coreweave.com/products/sunk/manage_sunk/shared-storage.md): Create and mount Persistent Volume Claims as shared storage across Slurm compute and login nodes on SUNK Standard - [Node-local storage and temporary files on Slurm nodes](https://docs.coreweave.com/products/sunk/manage_sunk/node-local-storage-and-tmp.md): What backs /tmp on SUNK Slurm nodes and how to keep temporary files on NVMe instead of RAM ##### ノードとジョブ - [Monitor Slurm node states](https://docs.coreweave.com/products/sunk/manage_sunk/slurm-node-states.md): Learn about the different states a Slurm node can be in, and how to check the state of a node. - [Monitor Slurm job states](https://docs.coreweave.com/products/sunk/manage_sunk/slurm-job-states.md): View real-time and historic information about Slurm jobs - [Drain and undrain Slurm nodes](https://docs.coreweave.com/products/sunk/manage_sunk/drain-and-undrain-nodes.md): Identify the reason a node was drained, manually drain and undrain nodes - [Use sidecar containers with SUNK](https://docs.coreweave.com/products/sunk/manage_sunk/add-sidecars-to-slurm-nodes.md): Add sidecar containers to Slurm compute and login node pods - [Set up AppArmor profile for Enroot](https://docs.coreweave.com/products/sunk/manage_sunk/enroot-apparmor.md): Configure the recommended AppArmor profile for Enroot unprivileged containers on SUNK compute nodes ###### ノードのトラブルシューティング - [Address nodes in the INVAL state](https://docs.coreweave.com/products/sunk/manage_sunk/inval-node-state.md): Troubleshoot Slurm nodes in the INVAL state caused by configuration mismatches between slurm.conf and slurmd - [Address nodes stuck during cleanup operations](https://docs.coreweave.com/products/sunk/manage_sunk/stuck-slurm-nodes.md): Troubleshooting steps for Slurm nodes stuck in the Completing state #### ワークロードを実行する - [Use Docker in SUNK](https://docs.coreweave.com/products/sunk/run_workloads/docker-in-docker.md): Enable Docker-in-Docker inside SUNK clusters by configuring AppArmor, privileged pods, and s6 services - [Run custom scripts with s6](https://docs.coreweave.com/products/sunk/run_workloads/run-scripts-with-s6.md): Run custom longrun and oneshot scripts on SUNK compute and login nodes using the s6 service manager ##### Pod スケジューラー - [Schedule Kubernetes Pods with the SUNK Pod Scheduler](https://docs.coreweave.com/products/sunk/run_workloads/schedule-kubernetes-pods.md): Schedule Kubernetes Pods alongside Slurm jobs using the SUNK Pod Scheduler - [Run workloads with gang scheduling](https://docs.coreweave.com/products/sunk/run_workloads/gang-scheduling.md): Schedule Kubernetes Jobs and LeaderWorkerSet groups with shared Slurm allocations in SUNK - [Manage resources with the SUNK Pod Scheduler](https://docs.coreweave.com/products/sunk/run_workloads/manage-scheduler-resources.md): Configure Node sharing, resource allocation, and GPU management for Kubernetes Pods scheduled through Slurm - [Static CPU allocation and the SUNK Pod Scheduler](https://docs.coreweave.com/products/sunk/run_workloads/static-cpu-allocation.md): Why Kubernetes static CPU allocation is incompatible with the SUNK Pod Scheduler and how to prevent node drains. ##### Prolog と Epilog - [Run Prolog and Epilog scripts on SUNK](https://docs.coreweave.com/products/sunk/run_workloads/prolog-epilog.md): Overview of Slurm Prolog and Epilog scripts in SUNK, including how they work and failure handling behavior - [Create and configure Prolog and Epilog scripts](https://docs.coreweave.com/products/sunk/run_workloads/create-prolog-epilog.md): Create Kubernetes ConfigMaps with Prolog and Epilog scripts and configure SUNK to run them with Slurm jobs ##### Slurm ジョブ - [Avoid stranding GPUs with CPU and memory requests](https://docs.coreweave.com/products/sunk/run_workloads/avoid-stranded-gpus.md): Size --cpus-per-task and --mem-per-cpu so Slurm jobs use every GPU on a SUNK node instead of stranding one. - [Handle Slurm signals for graceful shutdown](https://docs.coreweave.com/products/sunk/run_workloads/handle-slurm-signals.md): Trap Slurm termination signals to checkpoint training jobs and report an accurate run status instead of a crash - [Checkpoint and restart Slurm jobs after a node failure](https://docs.coreweave.com/products/sunk/run_workloads/checkpoint-and-restart.md): Configure a Slurm job on SUNK to automatically requeue and resume from checkpoint after a node failure - [Use scrontab to schedule recurring batch jobs](https://docs.coreweave.com/products/sunk/run_workloads/using-scrontab.md): Schedule recurring batch jobs in SUNK using scrontab, a cron-like Slurm utility for automated job submission - [Run jobs on Grace CPU instances](https://docs.coreweave.com/products/sunk/run_workloads/run-jobs-on-grace-cpu.md): Run SUNK jobs on Arm-based Grace CPU instances (GB200, GB300, GH200), including page-size and binary compatibility - [Job priority, QoS, and preemption](https://docs.coreweave.com/products/sunk/run_workloads/job-priority-qos-preemption.md): On SUNK, priority weights and preemption start unset; understand the model before you tune it ##### ワークロードを最適化する - [Enable GPU straggler detection](https://docs.coreweave.com/products/sunk/optimize_workloads/enable-gpu-straggler-detection.md): Enable the GPU Straggler Detection plugin in SUNK and confirm it loads, so you can find GPUs that fall behind. - [Topology and block scheduling in Slurm](https://docs.coreweave.com/products/sunk/optimize_workloads/topology-scheduling.md): Configure topology and block scheduling in Slurm for optimized job placement on NVLink-connected GPU systems - [Manage memory with cache-dropper](https://docs.coreweave.com/products/sunk/optimize_workloads/cache-dropper.md): Use cache-dropper to manage memory between exclusive Slurm jobs on a node - [Create custom images for SUNK Standard](https://docs.coreweave.com/products/sunk/optimize_workloads/custom-images.md): Customize published SUNK images with Dockerfiles and configure them for SUNK Standard compute and login nodes - [Manage resource binding with task plugins](https://docs.coreweave.com/products/sunk/optimize_workloads/task-plugins.md): Optimize performance by binding tasks to common resources within a node - [Customize cluster behavior with Lua plugins on SUNK Standard](https://docs.coreweave.com/products/sunk/optimize_workloads/lua-plugins.md): Use Slurm's custom job submit script to enforce rules and control job submissions on SUNK Standard clusters - [Optimize CPU binding and NUMA affinity for multi-GPU jobs](https://docs.coreweave.com/products/sunk/optimize_workloads/cpu-binding-numa-affinity.md): Fix Slurm cyclic task distribution that causes NUMA misalignment and slows multi-GPU jobs on SUNK - [Measure MFU and job performance](https://docs.coreweave.com/products/sunk/optimize_workloads/measuring-mfu-and-job-performance.md): Compute model FLOPs utilization, interpret DCGM GPU metrics, and tell a tuning problem from an infrastructure problem - [Diagnose unexplained memory pressure on a SUNK node](https://docs.coreweave.com/products/sunk/optimize_workloads/memory-accounting.md): Tell reclaimable page cache from pinned shmem and system overhead when Pod metrics look fine but the node OOMs #### リファレンス - [SUNK reference](https://docs.coreweave.com/products/sunk/reference.md): Container images, custom resource fields, configuration parameters, and version mappings for SUNK. - [SUNK to Slurm version mapping](https://docs.coreweave.com/products/sunk/reference/sunk-slurm-versions.md): Mapping table between SUNK release versions and their corresponding Slurm versions - [SCIM parameter reference](https://docs.coreweave.com/products/sunk/reference/nsscache_reference.md): Complete reference of configurable SCIM parameters for SUNK user identity management ##### SUNK - [SUNK parameter reference](https://docs.coreweave.com/products/sunk/reference/sunk-parameters.md): Complete parameter reference for the SUNK operator Helm chart, including image, MySQL, and NVIDIA plugin settings - [SunkCluster reference](https://docs.coreweave.com/products/sunk/reference/sunkcluster-reference.md): Reference parameters for the SunkCluster custom resource that defines a SUNK Self-Service cluster ##### Slurm - [Slurm images](https://docs.coreweave.com/products/sunk/reference/slurm-images.md): Reference for SUNK-built Slurm container images, including base and extras variants with supported CUDA versions - [Slurm parameter reference](https://docs.coreweave.com/products/sunk/reference/slurm-parameters.md): Complete parameter reference for the Slurm Helm chart, including accounting, compute, and deployment settings - [Slurm login parameter reference](https://docs.coreweave.com/products/sunk/reference/slurm-login-parameters.md): Complete parameter reference for the Slurm login Helm chart, including directory cache and affinity settings ### CoreWeave Inference - [About CoreWeave Inference](https://docs.coreweave.com/products/inference.md): Compare Serverless Inference, Dedicated Inference, and Inference on CKS to choose how to serve your models #### Dedicated Inference - [About Dedicated Inference](https://docs.coreweave.com/products/inference/dedicated.md): Deploy your own model weights on dedicated GPU infrastructure with full control over gateways, scaling, and capacity - [Getting started with Dedicated Inference](https://docs.coreweave.com/products/inference/getting-started.md): Create a gateway, deploy a model, and send your first Dedicated Inference request - [Deploy Dedicated Inference with Terraform](https://docs.coreweave.com/products/inference/deploy-with-terraform.md): Manage Dedicated Inference gateways and deployments as code with the CoreWeave Terraform provider - [RL Rollouts](https://docs.coreweave.com/products/inference/rl-rollouts.md): Generate reinforcement learning rollouts and load updated policy weights into a running deployment #### CKS 上の Inference - [About inference on CKS](https://docs.coreweave.com/products/inference/cks.md): Run self-managed inference workloads on CoreWeave Kubernetes Service #### リファレンス ##### Dedicated Inference API - [Dedicated Inference API reference](https://docs.coreweave.com/products/inference/reference/api-overview.md): Authentication, protocols, status values, and error formats for the Dedicated Inference API ###### CapacityClaimService - [List capacity claims](https://docs.coreweave.com/ja/products/inference/reference/api-overview/capacityclaimservice/list-capacity-claims.md): Fetch every CapacityClaim the caller can read, with optional `updatedAfter` filter. - [Create capacity claim](https://docs.coreweave.com/ja/products/inference/reference/api-overview/capacityclaimservice/create-capacity-claim.md): Reserve GPU instances for inference deployments via a new CapacityClaim. - [Get capacity claim parameters](https://docs.coreweave.com/ja/products/inference/reference/api-overview/capacityclaimservice/get-capacity-claim-parameters.md): Enumerate the available instance types per Availability Zone for CapacityClaim creation. - [Get capacity claim](https://docs.coreweave.com/ja/products/inference/reference/api-overview/capacityclaimservice/get-capacity-claim.md): Fetch a single CapacityClaim by its UUID. - [Delete capacity claim](https://docs.coreweave.com/ja/products/inference/reference/api-overview/capacityclaimservice/delete-capacity-claim.md): Delete a CoreWeave CapacityClaim by its UUID. - [Update capacity claim](https://docs.coreweave.com/ja/products/inference/reference/api-overview/capacityclaimservice/update-capacity-claim.md): Update the resources reserved by a CapacityClaim. ###### DeploymentService - [List deployments](https://docs.coreweave.com/ja/products/inference/reference/api-overview/deploymentservice/list-deployments.md): Fetch every deployment the caller can read, with optional gateway and `updatedAfter` filters. - [Create deployment](https://docs.coreweave.com/ja/products/inference/reference/api-overview/deploymentservice/create-deployment.md): Create a new CoreWeave inference model deployment. - [Get deployment parameters](https://docs.coreweave.com/ja/products/inference/reference/api-overview/deploymentservice/get-deployment-parameters.md): Enumerate the available gateways, runtime engines, and instance types for deployment creation. - [Get deployment](https://docs.coreweave.com/ja/products/inference/reference/api-overview/deploymentservice/get-deployment.md): Fetch a single deployment by its ID. - [Delete deployment](https://docs.coreweave.com/ja/products/inference/reference/api-overview/deploymentservice/delete-deployment.md): Delete a CoreWeave inference deployment by its ID. - [Update deployment](https://docs.coreweave.com/ja/products/inference/reference/api-overview/deploymentservice/update-deployment.md): Update the spec of a CoreWeave inference deployment. ###### GatewayService - [List gateways](https://docs.coreweave.com/ja/products/inference/reference/api-overview/gatewayservice/list-gateways.md): Fetch every inference gateway the caller can read, with optional `updatedAfter` filter. - [Create gateway](https://docs.coreweave.com/ja/products/inference/reference/api-overview/gatewayservice/create-gateway.md): Create a CoreWeave inference gateway with one authentication type and one routing strategy. - [Get gateway parameters](https://docs.coreweave.com/ja/products/inference/reference/api-overview/gatewayservice/get-gateway-parameters.md): Enumerate the available Availability Zones for gateway creation. - [Get gateway](https://docs.coreweave.com/ja/products/inference/reference/api-overview/gatewayservice/get-gateway.md): Fetch a single inference gateway by its UUID. - [Delete gateway](https://docs.coreweave.com/ja/products/inference/reference/api-overview/gatewayservice/delete-gateway.md): Delete a CoreWeave inference gateway by its UUID. - [Update gateway](https://docs.coreweave.com/ja/products/inference/reference/api-overview/gatewayservice/update-gateway.md): Update the spec of a CoreWeave inference gateway. ###### HotLoadService - [List hot-loads](https://docs.coreweave.com/ja/products/inference/reference/api-overview/hotloadservice/list-hot-loads.md): Fetch the caller's checkpoint hot-loads, newest first, with an optional deployment filter. - [Create hot-load](https://docs.coreweave.com/ja/products/inference/reference/api-overview/hotloadservice/create-hot-load.md): Load a full or delta checkpoint into a running deployment's replicas. - [List hot-load replicas](https://docs.coreweave.com/ja/products/inference/reference/api-overview/hotloadservice/list-hot-load-replicas.md): Read per-replica progress for a hot-load, including the checkpoint each replica serves. - [Get hot-load](https://docs.coreweave.com/ja/products/inference/reference/api-overview/hotloadservice/get-hot-load.md): Read one checkpoint hot-load's specification, snapshot chain, and aggregate state. - [Cancel hot-load](https://docs.coreweave.com/ja/products/inference/reference/api-overview/hotloadservice/cancel-hot-load.md): Cancel an in-flight checkpoint hot-load. Not yet implemented. #### チュートリアル - [Deploy Kimi K3 on Dedicated Inference](https://docs.coreweave.com/products/inference/tutorials/deploy-kimi-k3.md): Serve Moonshot AI's Kimi K3 on GB300 GPUs with expert parallelism and speculative decoding #### インテグレーション - [Connect Claude Code and Claude Desktop to CoreWeave Inference](https://docs.coreweave.com/products/inference/connect-claude-code-and-desktop.md): Point Claude Code and Claude Desktop at CoreWeave Serverless or Dedicated Inference through a translation proxy #### 概念 - [Models and deployments](https://docs.coreweave.com/products/inference/models.md): Configure model deployments with custom weights, GPU selection, and inference runtimes - [Gateways](https://docs.coreweave.com/products/inference/gateways.md): How gateways provide authentication, routing, and traffic management for inference deployments - [Scaling](https://docs.coreweave.com/products/inference/scaling.md): Configure autoscaling and reserve capacity for inference deployments - [Speculative decoding](https://docs.coreweave.com/products/inference/speculative-decoding.md): Speculate tokens with a draft model to cut decode latency on a Dedicated Inference deployment - [Billing](https://docs.coreweave.com/products/inference/billing.md): How Dedicated Inference is billed and how to optimize inference costs ### CoreWeave Sandbox - [About CoreWeave Sandbox](https://docs.coreweave.com/products/sandboxes.md): Run agent tools, evaluations, and training workloads in isolated environments on CoreWeave. - [Get started with serverless sandboxes](https://docs.coreweave.com/products/sandboxes/get-started.md): Run your first serverless sandbox with a CoreWeave API access token. - [Use your own compute](https://docs.coreweave.com/products/sandboxes/get-started-cks.md): Enable sandboxes on your CoreWeave Kubernetes Service cluster and run your first command. - [Support](https://docs.coreweave.com/redirects/sandboxes/support.md) #### 概念 - [Sandboxes architecture](https://docs.coreweave.com/products/sandboxes/architecture.md): How runners, policies, and the gateway fit together across one or more CKS clusters. - [Sandbox lifecycle](https://docs.coreweave.com/products/sandboxes/client/guides/sandbox-lifecycle.md): Understand sandbox states, creation patterns, waiting, and shutdown behavior. - [Sync versus async patterns](https://docs.coreweave.com/products/sandboxes/client/guides/sync-vs-async.md): Choose between synchronous and asynchronous patterns in the CoreWeave sandbox Python SDK. #### チュートリアル - [Configuration](https://docs.coreweave.com/products/sandboxes/client/tutorial/configuration.md): Configure sandbox resources, images, and environment settings. - [Run commands in a sandbox environment](https://docs.coreweave.com/products/sandboxes/client/tutorial/running-commands.md): Execute commands inside a sandbox and capture output. - [Stream real-time output from a sandbox](https://docs.coreweave.com/products/sandboxes/client/tutorial/streaming.md): Stream command output in real time from a sandbox. - [File operations](https://docs.coreweave.com/products/sandboxes/client/tutorial/files.md): Tutorial for uploading and downloading files to and from CoreWeave sandbox instances - [Multiple sandboxes](https://docs.coreweave.com/products/sandboxes/client/tutorial/multiple-sandboxes.md): Tutorial for managing multiple CoreWeave sandbox instances using sessions with shared defaults - [Remote functions](https://docs.coreweave.com/products/sandboxes/client/tutorial/remote-functions.md): Run Python functions inside sandboxes with automatic serialization. - [Cleanup](https://docs.coreweave.com/products/sandboxes/client/tutorial/cleanup.md): Terminate sandboxes and manage cleanup with context managers, explicit stops, and global atexit handlers. - [Run pull request checks in sandboxes](https://docs.coreweave.com/products/sandboxes/tutorials/github-actions.md): Run repository tests in CoreWeave sandboxes from GitHub Actions, with optional advisory AI review. - [Run a computer-use agent in a sandbox](https://docs.coreweave.com/products/sandboxes/tutorials/computer-use.md): Run a Linux desktop in a serverless sandbox and let an agent operate it through screenshots and input. - [RL training](https://docs.coreweave.com/products/sandboxes/client/guides/rl-training.md): Use sandboxes for reinforcement learning training environments. #### サンドボックスを使用する ##### エージェントを統合する - [Run agents on Sandboxes](https://docs.coreweave.com/products/sandboxes/agents.md): Run interactive and autonomous agents with longer sessions, parallel work, and tools on CoreWeave. - [Claude Code](https://docs.coreweave.com/products/sandboxes/agents/claude-code.md): Run Claude Code in a CoreWeave sandbox and interact from your terminal, Claude apps, or claude.ai. - [Run Claude Code cloud sessions](https://docs.coreweave.com/products/sandboxes/agents/claude-code-cloud.md): Deploy a Claude Code orchestrator that creates a CoreWeave sandbox for each cloud session. - [Run OpenCode in a Sandbox](https://docs.coreweave.com/products/sandboxes/agents/opencode.md): Run OpenCode in a sandbox with W&B Serverless Inference or another model provider. - [Run Muse Code in a Sandbox](https://docs.coreweave.com/products/sandboxes/agents/muse-code.md): Run Muse Code CLI on a repository in a sandbox, retrieve the result, and stop compute. - [Run Pi in a Sandbox](https://docs.coreweave.com/products/sandboxes/agents/pi.md): Run Pi from an attached terminal with optional W&B Serverless Inference, then retrieve your work. - [Use Sandboxes with Claude Managed Agents](https://docs.coreweave.com/products/sandboxes/agents/claude-managed-agents.md): Connect a self-hosted sandbox worker to Claude Managed Agents and verify tool execution on CoreWeave. - [Run OpenAI Agents API on Sandboxes](https://docs.coreweave.com/products/sandboxes/agents/openai-agents-api.md): Connect OpenAI-managed agents to a CoreWeave sandbox for tool execution and shared-workspace analysis. - [Run Codex in a Sandbox](https://docs.coreweave.com/products/sandboxes/agents/codex.md): Run Codex CLI in a sandbox, connect a local CLI, or use desktop and mobile apps with a remote project. - [Run Cursor in a Sandbox](https://docs.coreweave.com/products/sandboxes/agents/cursor.md): Run Cursor CLI interactively or automate coding tasks in a CoreWeave sandbox. - [Run Devin Outposts on Sandboxes](https://docs.coreweave.com/products/sandboxes/agents/devin-outposts.md): Let Devin plan in Devin Cloud while its commands run in a CoreWeave sandbox you control. ##### 評価 - [Run evaluations on CoreWeave sandboxes](https://docs.coreweave.com/products/sandboxes/evals.md): Choose a benchmark workflow to evaluate code patches, coding agents, and tool-calling models in sandboxes. - [Run SWE-bench evaluations](https://docs.coreweave.com/products/sandboxes/evals/swe-bench.md): Run SWE-bench evaluations using CoreWeave sandboxes. - [Run DeepSWE v1.1 evaluations](https://docs.coreweave.com/products/sandboxes/evals/deepswe.md): Generate code patches and grade them in separate sandboxes with a small DeepSWE v1.1 sample. - [Evaluate models with AutomationBench](https://docs.coreweave.com/products/sandboxes/evals/automationbench.md): Run AutomationBench with Harbor, CoreWeave Sandboxes, and a configurable model endpoint. ##### 設定する - [Sandbox configuration](https://docs.coreweave.com/products/sandboxes/client/guides/sandbox-configuration.md): Configure sandbox resources, images, networking, timeouts, and storage. - [Sandbox placement and spillover](https://docs.coreweave.com/products/sandboxes/placement.md): Compare serverless and CKS sandboxes, select where they run, and configure fallback between modes. - [Run a GPU sandbox](https://docs.coreweave.com/products/sandboxes/gpu-sandboxes.md): Request GPUs for a sandbox on serverless capacity, which is virtualized, or on your own CKS cluster, which isn't. - [Sessions](https://docs.coreweave.com/products/sandboxes/client/guides/sessions.md): Group and manage multiple sandboxes with sessions. - [Sandbox templates](https://docs.coreweave.com/products/sandboxes/profiles/templates.md): Create reusable sandbox templates, start sandboxes from them, and attach admin pod settings on CKS clusters. - [Create a sandbox from a Compose file](https://docs.coreweave.com/products/sandboxes/create-from-file.md): Start a multi-container sandbox by uploading a Docker Compose file to the sandbox API. - [Implement a warm pool](https://docs.coreweave.com/products/sandboxes/warm-pools.md): Keep sandboxes ready for incoming work using a Python warm pool recipe. - [Run long-running sandboxes](https://docs.coreweave.com/products/sandboxes/long-running-sandboxes.md): Configure sandbox lifetimes up to 30 days with Serverless or CKS placement. - [Cleanup patterns](https://docs.coreweave.com/products/sandboxes/client/guides/cleanup-patterns.md): Patterns for reliable sandbox cleanup and resource management. - [Discover available runners](https://docs.coreweave.com/products/sandboxes/client/guides/discovery.md): List the runners in your organization, read the hardware they advertise, and check free capacity. ###### 環境変数 - [Environment variables](https://docs.coreweave.com/products/sandboxes/client/guides/environment-variables.md): Set and manage environment variables in CWSandbox instances, including session-level defaults - [Environment variables](https://docs.coreweave.com/products/sandboxes/environment-variables.md): Environment variables CoreWeave sets inside every sandbox, and how to override them. ##### コードを実行する - [Command execution](https://docs.coreweave.com/products/sandboxes/client/guides/execution.md): Run commands in sandboxes with streaming, process control, and error handling. - [Interactive shells and TTY](https://docs.coreweave.com/products/sandboxes/client/guides/interactive-shells.md): Use interactive TTY sessions with shell() for terminal applications and raw byte streaming. - [Remote functions](https://docs.coreweave.com/products/sandboxes/client/guides/remote-functions.md): Execute Python functions inside sandboxes with the function decorator. - [Sandbox logging](https://docs.coreweave.com/products/sandboxes/client/guides/logging.md): Stream container logs (stdout and stderr) from a sandbox's main process with stream_logs(). ##### ファイルとストレージ - [File operations](https://docs.coreweave.com/products/sandboxes/client/guides/file-operations.md): Read, write, and manage files within CoreWeave sandbox instances using the CWSandbox client - [Volumes](https://docs.coreweave.com/products/sandboxes/volumes.md): Give a sandbox writable scratch space, or mount a registered volume that several sandboxes share. - [Grant sandboxes access to AI Object Storage](https://docs.coreweave.com/products/sandboxes/operations/object-storage-access.md): Let sandboxes exchange their built-in OIDC token for temporary AI Object Storage credentials ###### ファイルシステムスナップショット - [File system snapshots](https://docs.coreweave.com/products/sandboxes/file-system-snapshots.md): Snapshot a sandbox's working directory to object storage, then restore or fork it into new sandboxes. - [Bring your own bucket for file system snapshots](https://docs.coreweave.com/products/sandboxes/file-system-snapshots/bring-your-own-bucket.md): Store sandbox file system snapshots in an object storage bucket you own instead of the CoreWeave-managed default. ##### ネットワーク - [Public endpoints](https://docs.coreweave.com/products/sandboxes/public-endpoints.md): Give a service inside a sandbox a public address, with platform TLS or TLS passthrough. - [Direct sandbox connections](https://docs.coreweave.com/products/sandboxes/client/guides/direct-connections.md): Route exec, log, and file operations over a direct mTLS connection instead of the gateway. #### サンドボックスを管理する - [Deploy and manage a runner](https://docs.coreweave.com/products/sandboxes/operations/managed-runners.md): Deploy a runner on a CKS cluster, check its health, update its configuration, and delete it. - [Configure Node allocation budgets](https://docs.coreweave.com/products/sandboxes/operations/node-allocation-budgets.md): Limit sandbox CPU and memory requests across all Nodes, selected Node Pools, or groups of Nodes - [Integrate sandboxes with the SUNK Pod Scheduler](https://docs.coreweave.com/products/sandboxes/profiles/sunk-integration.md): Route sandbox pods through the SUNK Pod Scheduler so Slurm places them alongside your Slurm jobs. - [Monitor sandbox usage metrics](https://docs.coreweave.com/products/sandboxes/operations/usage-metrics.md): Query per-sandbox lifecycle and usage metrics for CoreWeave Sandboxes with PromQL. ##### ポリシー - [Policies overview](https://docs.coreweave.com/products/sandboxes/profiles/profiles.md): How a sandbox resolves against the policy that governs its cluster, and what a policy can constrain. - [Configure a sandbox policy](https://docs.coreweave.com/products/sandboxes/profiles/configure.md): Set the constraints that bound what sandboxes on a cluster may request, and the defaults they inherit. - [Policy examples](https://docs.coreweave.com/products/sandboxes/profiles/profile-examples.md): Complete policy documents for five common cluster postures, for data science, agents, CI, and background workers. - [Policy reference](https://docs.coreweave.com/products/sandboxes/reference/profile.md): Every field in a sandbox policy, with types, defaults, and validation rules. #### リファレンス - [Limits and quotas](https://docs.coreweave.com/products/sandboxes/reference/limits-and-quotas.md): Plan sandbox concurrency and resources, understand lifetime limits, and handle quota errors. ##### API リファレンス - [Sandbox Python client](https://docs.coreweave.com/products/sandboxes/client.md): Python SDK for creating and managing CoreWeave sandbox environments. ###### コア - [Sandbox](https://docs.coreweave.com/products/sandboxes/client/ref/core/sandbox.md): CWSandbox client with sync/async hybrid API. - [Session](https://docs.coreweave.com/products/sandboxes/client/ref/core/session.md): Manages sandbox lifecycle and provides function execution. - [SandboxDefaults](https://docs.coreweave.com/products/sandboxes/client/ref/core/sandbox-defaults.md): Immutable configuration defaults for sandbox creation. ###### 型 - [OperationRef](https://docs.coreweave.com/products/sandboxes/client/ref/types/operation-ref.md): Generic ref for async operations with lazy result retrieval. - [Process](https://docs.coreweave.com/products/sandboxes/client/ref/types/process.md): Handle for a running process with streaming stdout/stderr. - [ProcessResult](https://docs.coreweave.com/products/sandboxes/client/ref/types/process-result.md): Result from a completed streaming exec operation. - [StreamReader](https://docs.coreweave.com/products/sandboxes/client/ref/types/stream-reader.md): Sync and async iterable stream reader. - [StreamWriter](https://docs.coreweave.com/products/sandboxes/client/ref/types/stream-writer.md): Sync and async writer for streaming input to a process. - [RemoteFunction](https://docs.coreweave.com/products/sandboxes/client/ref/types/remote-function.md): Wrapper for remote function execution in sandboxes. - [TerminalSession](https://docs.coreweave.com/products/sandboxes/client/ref/types/terminal-session.md): Handle for an interactive TTY session in a sandbox. - [TerminalResult](https://docs.coreweave.com/products/sandboxes/client/ref/types/terminal-result.md): Result from a completed terminal session. - [SandboxFileType](https://docs.coreweave.com/products/sandboxes/client/ref/types/sandbox-file-type.md): Document type for ``Sandbox.run_from_file``. - [FieldViolation](https://docs.coreweave.com/products/sandboxes/client/ref/types/field-violation.md): Field-level validation failure returned by the backend. ###### 設定 - [ResourceOptions](https://docs.coreweave.com/products/sandboxes/client/ref/configuration/resources.md): Resource configuration for sandbox CPU, memory, and GPU. - [NetworkOptions](https://docs.coreweave.com/products/sandboxes/client/ref/configuration/network-options.md): Network deny flags and create-time ingress/egress grants. - [Secret](https://docs.coreweave.com/products/sandboxes/client/ref/configuration/secret.md): A secret to inject from a store into a sandbox environment variable. - [SandboxStatus](https://docs.coreweave.com/products/sandboxes/client/ref/configuration/sandbox-status.md): Sandbox lifecycle status values. - [AuthStrategy](https://docs.coreweave.com/products/sandboxes/client/ref/configuration/auth-strategy.md): Built-in authentication strategies. - [AuthProvider](https://docs.coreweave.com/products/sandboxes/client/ref/configuration/auth-provider.md): Provider that resolves authentication for a Sandbox API endpoint. - [AuthHeaders](https://docs.coreweave.com/products/sandboxes/client/ref/configuration/auth-headers.md): Resolved authentication headers and strategy used. - [set_auth_mode](https://docs.coreweave.com/products/sandboxes/client/ref/configuration/set-auth-mode.md): Set the legacy process-global auth mode. ###### ワークロード - [Containers](https://docs.coreweave.com/products/sandboxes/client/ref/workload/containers.md): Container definitions, observed container state, and per-container privilege. - [Placement](https://docs.coreweave.com/products/sandboxes/client/ref/workload/placement.md): Where a sandbox runs and how its data operations travel. - [Object storage access](https://docs.coreweave.com/products/sandboxes/client/ref/workload/object-storage.md): Temporary object-storage credentials injected into a sandbox's containers. ###### ネットワーク - [Network rules](https://docs.coreweave.com/products/sandboxes/client/ref/networking/network-rules.md): Create-time egress and ingress rules, and the selectors they accept. - [Services](https://docs.coreweave.com/products/sandboxes/client/ref/networking/services.md): Typed service ports a sandbox exposes. - [Endpoints](https://docs.coreweave.com/products/sandboxes/client/ref/networking/endpoints.md): Product endpoints on public services, and the status shapes echoed back. ###### ボリューム - [Volumes](https://docs.coreweave.com/products/sandboxes/client/ref/volumes/volumes.md): Registered volumes, scratch volumes, and how they mount into containers. - [Types](https://docs.coreweave.com/products/sandboxes/client/ref/volumes/types.md): Enumerations describing volume state, locality, and backing store. ###### ファイルシステムスナップショット - [Snapshots](https://docs.coreweave.com/products/sandboxes/client/ref/snapshots/snapshots.md): File-system snapshot records, mount options, and archive bucket configuration. - [Types](https://docs.coreweave.com/products/sandboxes/client/ref/snapshots/types.md): Enumerations describing snapshot status, trigger, and bucket ownership. ###### ディスカバリー - [Functions](https://docs.coreweave.com/products/sandboxes/client/ref/discovery/functions.md): Query the runners visible to your organization. - [Types](https://docs.coreweave.com/products/sandboxes/client/ref/discovery/types.md): Data types returned by discovery functions. ###### 例外 - [Exceptions](https://docs.coreweave.com/products/sandboxes/client/ref/exceptions/exceptions.md): Exception hierarchy for cwsandbox error handling. ###### ユーティリティ - [results](https://docs.coreweave.com/products/sandboxes/client/ref/utilities/results.md): Block for one or more OperationRefs and return results. - [wait](https://docs.coreweave.com/products/sandboxes/client/ref/utilities/wait.md): Wait for waitables to complete, return (done, pending). ##### コントロールプレーン - [API overview](https://docs.coreweave.com/products/sandboxes/reference/control-plane-api.md): Base URL, authentication, versioning, pagination, and error handling for the CoreWeave Sandbox control plane API. - [Protobuf](https://docs.coreweave.com/products/sandboxes/reference/protos.md): Find the CoreWeave Sandbox Protobuf definitions, the host that serves each service, and the generated SDKs. ## OpenAPI Specs - [openapi](/openapi/inference/openapi.yaml)